On Tue, Nov 15, 2011 at 18:25, Vincent Massol <[email protected]> wrote:
> > On Nov 15, 2011, at 5:29 PM, Denis Gervalle wrote: > > > On Tue, Nov 15, 2011 at 17:11, Vincent Massol <[email protected]> > wrote: > > > >> > >> On Nov 15, 2011, at 5:03 PM, Denis Gervalle wrote: > >> > >>> Hi Devs, > >>> > >>> The implementation of the immutable version of reference is almost > ready > >>> now. It introduce the parameters on reference as suggested, but we now > >> have > >>> a discussion on how the constructors of "typed" entity reference should > >> be. > >>> > >>> My initial dev was to provide constructor like: > >> > >> you're missing something here :) > >> > >>> But Vincent have different vision of this, here its comment extracted > >> from > >> > >> I don't have a different vision. It's just that you limited your > proposal > >> to just Locale which clearly isn't good enough. > >> > >>> GitHub ( > >>> > >> > https://github.com/xwiki/xwiki-platform/commit/cea424914f40ce924afbc49b3159bc8934251aac#commitcomment-721603 > >> ) > >>> : > >>> > >>> My proposal was: > >>>> - generic params in EntityReference > >>>> - helpers methods for get/setLocale and get/setVersion in > >> DocumentReference > >>>> > >>>> Now the fact that we're making the refs immutable changed this since > >> it's > >>>> no longer possible. > >>>> > >>>> I don't think multiplying the constructor signatures is a good idea. > >>>> > >>>> We could either have: > >>>> > >>>> public DocumentReference(String wikiName, List spaceNames, String > >>>> pageName, Map<String, Object> parameters) > >>>> > >>>> or > >>>> > >>>> public DocumentReference(String wikiName, List spaceNames, String > >>>> pageName, Pair<String, Object>... parameters) > >>>> > >>>> where Pair is > >>>> > >> > http://commons.apache.org/lang/api-3.0-beta/org/apache/commons/lang3/Pair.html > >>>> > >>>> Maybe this needs some discussion on the devs list rather than here to > >> make > >>>> sure everyone sees it? > >>>> > >>> > >>> I am myself not really happy with that since I dislike the idea that > >>> parameter are generic on "typed" references. > >>> Do not like either the idea to provide keys for creating a Map or > Pairs, > >>> since the implementation details that use Map should not be so exposed > >> IMO. > >>> > >>> There should not be so much parameter on a single "typed" reference, > >> > >> I don't understand this sentence. > >> > >> The Map is <String, Object>. > >> > > > > I would means that this should not cause so many additional constructors, > > if we list them individually. > > So your have not the same vision then I have :) > > See below. > > >>> and > >>> these should be easy to provide. Creating Maps in java is not fun in > >> syntax > >>> for that IMO, and is far too open. > >> > >> Sure but the goal here is not to redo java… > >> > > > > Not my goals, just want to be explicit and easy to use. > > > > > >> > >>> I had propose using overloaded constructor like > >>> > >>> public DocumentReference(String wikiName, List<String> spaceNames, > String > >>> pageName, Locale locale) > >> > >> Again, this doesn't work. It only works for a **single** parameter. It > >> doesn't work for multiple parameters. How do you specify the Locale or > some > >> unknown String param? > >> > > > > I simply provide the need constructor, no more. > > If you say this then you say that we don't need generic parameters for > References which is what started the discussion… > > Also what you say is not correct at all since we already know we need > Version which you didn't put. > Not yet implemented. Isn't this your proposal initially ?... my needs is Locale for now. > > Last, it means passing null to the constuctors when you don't use the > parameters which is really really bad IMO or you'll need to define lots of > various constructors which increase exponentially with the number of > parameters we support. > Probably, but personally I was not expect so many parameter on a given "typed" reference. > > However this won't even work since you won't be able to support adding new > parameters. Users of the Reference system will need to modify its source if > they wish to add a new parameter which again goes against the initial > proposal. > Not sure what you means here. > > >>> or if something more flexible should be used > >>> > >>> public DocumentReference(String wikiName, List<String> spaceNames, > String > >>> pageName, Object... parameters) > >>> > >>> where parameters is later interpreted based on object type and limited > to > >>> those used for a given typed reference. > >> > >> This doesn't work. If I have 2 parameters of type String, how do you map > >> them automatically? > >> > > > > I suggest not to have loosely typed parameters, but strong one, like > Locale > > and Version. > > I'm a big big big -1 on this. > > There's no reason to create a fake String when all you need is a string. > Same for int, long, Number, etc. > Same for also if you want to have several Locales or several Versions. > There is for me good reason not to use String, we have seen what happen with String, and this why we now use Reference. > > >> Also the goal is to have unknown parameters so how can you do a mapping > >> for something unknown? :) > >> > > > > That is clearly not my goal, why do you want unknown parameters on > Document > > Reference ? > > Maybe this is where we don't agree. This is the original idea: to have a > set of "unknown" parameters for extensibility (from the POV of the > Reference of course, from the POV of XWiki they are not unknown obviously > ;)). Entity Reference, which is the base of any reference have and support the unknown references that are well know from typed references. There is almost nothing except typing at the document reference level. > This means that it's up to the users of the References to decide what > parameters to put and it's up to the consumers to decide what parameters to > support. > Agreed at the Entity Reference level, but for me a Document Reference should not have arbitrary parameters. What is the UC for this ? Since all parameters change the identity of a reference, I do not understand what you expect from arbitrary parameters except more issues. > > It seems to me your vision is to: > * Not support arbitrary parameters > for typed references ( like DocumentReference) but EntityReference does not have any a particular knowledge. > * Only support Version and Locale > currently for DocumentReference, but what else do you expect here ? I means that I do not expect to have to much parameters for a given typed reference for not supporting them in constructors has individual arguments. > * Provide 3 constructors, one where there's no Locale no Version, one with > Locale no version and one with Version on Locale > true, practical, documentary, and clearly what we really need right now. Whereas my vision is: > * Support arbitrary parameters for extensibility (like a URL if you prefer > which support arbitrary parameters) > * Have some "well-known" parameters so that xwiki code that users > References know what they can put as reference parameters and what they can > extract from XWiki parameters. > How could well-know it could be ? Do you means that you will later check the maps to validate the list of key allowed in a document ? I am not against this, but I am against forcing the creation of a map just to specify a language or a version. > > Thanks > -Vincent > > >> Thanks > >> -Vincent > >> > >>> Here Vincent comments on this: > >>> > >>>> The automatic mapping idea seems a bad idea to me (too magic and > doesn't > >>>> work in a lot of cases). > >>> > >>> > >>> Maybe some of you have an even better idea ? > >>> > >>> Denis > _______________________________________________ > devs mailing list > [email protected] > http://lists.xwiki.org/mailman/listinfo/devs > -- Denis Gervalle SOFTEC sa - CEO eGuilde sarl - CTO _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
